Blender's "Cycles X" Showing Nice Performance But Dropping OpenCL Support

[piotrj3]

Depends on how willing the Blender team would be to accept contributions of backends for:

• Radeon ProRender
• Intel Embree / OSPRay

One is willing to accept contributions, another is if that company (AMD/Intel) are willing to maintain it as well. Optix appeared in 2.8x and same 2 people work on Blender from Nvidia all time there, fact that cycles X works at all with OptiX is because he made contributions just few days before blog was posted. Nvidia didnt' make contribution and just left, they are all time there.

[Imout0]

I am supporting and suffering Blender because I thought it was a project that meant something. I don't see the point in suffering its myriad of incompatibilities, bugs and drawbacks if they are not staying "free". There are plenty of alternatives that do a far better job than Blender in every task.

[coder]

The developer who actually submitted OptiX render backend for new Cycles X, is literally Nvidia employee. As long as Nvidia contributes, and keeps making it work, I don't think anyone should protest for Nvidia contribution here.

Depends on how willing the Blender team would be to accept contributions of backends for:

• Radeon ProRender
• Intel Embree / OSPRay

[evil_core]

And to think coder and others berated me when I rightfully said on another thread that the market had rejected OpenCL for OneAPI and CUDA.

Yet another data point to show that.

AMD at this point should just immediately and whole sale abandon ROCm and move to OneAPI.

Codeplay has a version of OneAPI that runs on Nvidia.

Hell, even Fujitsu made an open sourced version of OneAPI to run on their ARM based Supercomputer Fugaku.

ROCm at this point is AMD's 3dNow! of GPU compute.

except AMD invented SYCL, ROCm provides SYCL and Intel OneApi is rebranded SYCL. RTFM next time!

[piotrj3]

The main Cycles developers are NVidia fanboys. So no surprise here. It was always "we implement it in CUDA, the rest is up to other people".

The developer who actually submitted OptiX render backend for new Cycles X, is literally Nvidia employee. As long as Nvidia contributes, and keeps making it work, I don't think anyone should protest for Nvidia contribution here.

[bridgman]

Sure. Just don't confuse this graphics thread for a mining thread.

Not sure I understand - the entire discussion in this thread has been about compute.

[alcalde]

Yes, but how many render engines are still supporting OpenCL into the future?

This is like WIndows and Microsoft Office. Ten years ago how many software programs were supporting Linux or how many companies would accept files that weren't MS Word?
If we don't support OpenCL, the proprietary crap wins. So we all HAVE to support OpenCL. Period. The same way we fought (and still fight) the uphill battle against Windows, MS Office, Photoshop, etc. The same way we BEAT Internet Explorer, proprietary browser plugins, flash, etc. We have to beat anyone who uses CUDA over the head, even if they're open source devs who have apparently gone mad, just as we had to register complaints when our banks insisted we "install their ActiveX plugin into Internet Explorer" to access our own bank accounts online and take our business elsewhere if necessary.

[alcalde]

For this moment, yes. Because OpenCL is crap anyways and only works reliable where CUDA also works reliable. Maybe Vulkan will fix this, maybe Radeon ProRender will also work in future. For now, there is only Nvidia anyways. At least on Linux.

That's like the people who used to say that you could only run Windows and use Microsoft Office because that's what everyone else was using.

"and only works reliable where CUDA also works reliable" No, it works quite reliably on AMD and Intel GPUs, unlike CUDA.

[MadeUpName]

We did that a few years ago... ROCm. Please don't confuse "we have not executed the entire roadmap yet" with "we do not have a roadmap".

Same ROCm stack for all recent GPUs on both Linux and Windows. That's it.

Sure. Just don't confuse this graphics thread for a mining thread.

[bridgman]

They should pick one direction and stick with it, publish a road map and then implement it..

We did that a few years ago... ROCm. Please don't confuse "we have not executed the entire roadmap yet" with "we do not have a roadmap".

Same ROCm stack for all recent GPUs on both Linux and Windows. That's it.

EDIT - for clarity I should mention that the ROCm stack includes multiple languages/APIs and that I expect OpenCL will continue to be one of those.